multicloudj-feature-dev

name	multicloudj-feature-dev
description	**REQUIRED for ANY feature development in multicloudj** - implements features end-to-end across client APIs, driver abstract classes, provider implementations (AWS/GCP/Ali), unit tests, and conformance tests with record/replay

MultiCloudJ End-to-End Feature Development

Overview

Implements a new feature across the multicloudj SDK through structured phases: requirements gathering via user interview, cross-cloud research for semantic uniformity, client/driver API design, provider implementations, unit tests, and conformance tests with WireMock record/replay.

Provider Isolation Principle

CRITICAL: Each provider implementation (AWS, GCP, Alibaba) is completely independent and must NEVER reference another provider in any way.

The Rule

Providers are isolated islands that:

• Each independently implements the abstract driver contract (e.g., AbstractBlobStore, AbstractDocstore)
• Have NO knowledge of other provider implementations
• Have NO dependencies on other provider modules
• Have NO comparisons or references to other providers in code OR comments

What This Means in Practice

NEVER:

• Import classes from another provider module (e.g., AwsBlobStore importing from blob-gcp)
• Add dependencies between provider modules in pom.xml
• Write comments like "Unlike GCP, AWS does X" or "Similar to Alibaba's approach"
• Write comments like "GCP handles this differently" or "AWS uses a different pattern"
• Copy code between providers with comments referencing the source
• Compare implementation approaches in javadoc or inline comments

DO:

• Each provider implements the driver contract independently
• Document WHY a provider does something based on that provider's SDK behavior
• Focus comments on the cloud provider's native SDK, not other multicloudj providers
• Let conformance tests verify behavioral parity, not implementation similarity

Example: WRONG ❌

// AWS implementation
@Override
protected PutObjectResponse doPutObject(PutObjectRequest request) {
    // Unlike GCP which uses resumable uploads, AWS uses multipart
    // Similar to how AliOss handles large objects
    return s3Client.putObject(...);
}

Example: CORRECT ✅

// AWS implementation
@Override
protected PutObjectResponse doPutObject(PutObjectRequest request) {
    // S3 automatically handles objects up to 5GB in a single PUT operation
    return s3Client.putObject(...);
}

Why Provider Isolation Matters

1. Maintainability: Each provider evolves independently based on its cloud SDK changes
2. Testability: Provider tests don't break when other providers change
3. Clarity: Implementation decisions are based on the provider's SDK, not other providers
4. Extensibility: New providers can be added without touching existing ones
5. Semantic Uniformity: The driver contract (abstract class) defines the unified behavior, not cross-provider comparisons

The driver contract and conformance tests ensure all providers behave the same for the end user. Providers achieve this independently, not by copying or comparing with each other.

When to Use

• Adding a new operation or capability to an existing service (blob, docstore, pubsub, sts)
• Adding a new service module end-to-end
• User says "add feature", "implement X operation", "support Y across providers"

When NOT to Use

• Bug fixes to existing implementations (use systematic-debugging)
• Refactoring without behavior change
• Documentation-only changes (use docs-guides)

Process Flow

flowchart TD
    A([User request received]) --> B[Phase 1: Interview]
    B --> C[Phase 2: Cross-cloud research]
    C --> D{Semantic differences?}
    D -->|yes| E[Interview user on resolution]
    D -->|no| F[Phase 3: API design]
    E --> F
    F --> G{User approves API?}
    G -->|yes| H[Phase 4: Provider implementations]
    G -->|revise| F
    H --> I[Phase 5: Unit tests]
    I --> J[Phase 6: Conformance tests]
    J --> K{Run record mode?}
    K -->|yes| L[Collect credentials and record]
    K -->|later| M[Write abstract IT only]
    L --> N[Phase 7: Create PR]
    M --> N
    N --> O([Done])

Phase 1: User Interview

Gather requirements before any code. Ask the user:

1. What service? (blob, docstore, pubsub, sts, or new)
2. What operation? (e.g., "object versioning", "batch delete with filters", "topic dead-letter queue")
3. What's the user-facing behavior? (what should the end-user SDK call look like?)
4. Which providers? (AWS/GCP/Ali - default is all three)
5. Are there known provider differences? (the user may already know gotchas)
6. Any constraints? (backwards compatibility, performance, specific SDK versions)

Do NOT proceed until you have clear answers to at least questions 1-3.

Phase 2: Cross-Cloud Research

Research how each cloud provider implements this feature natively. For each provider (AWS, GCP, Alibaba):

1. Read the cloud SDK documentation - understand the native API surface
2. Identify the SDK classes/methods that implement this feature
3. Note semantic differences between providers:
   • Different parameter names or types
   • Different behavior for edge cases (nulls, empty lists, limits)
   • Features supported by some providers but not others
   • Different error codes/conditions for the same failure

Present findings to user as a comparison table:

Aspect	AWS	GCP	Alibaba
API method	s3.putObjectRetention()	storage.objects().update()	oss.setBucketLifecycle()
Supports X	Yes	Partial	No
Error on Y	403	404	400

Resolving Semantic Differences

If providers differ, present the user with options:

• Option A: Lowest common denominator - only expose what all providers support
• Option B: Best-effort with capability flags - expose all, throw UnsupportedOperationException where unavailable
• Option C: Semantic translation - map different provider behaviors to a unified model

Interview the user on which approach to take. The goal is semantic uniformity for the end user - they should not need to know which cloud they're running on.

Phase 3: API Design

Design the cloud-agnostic API. This involves changes to the -client module:

3a. Request/Response Objects

Create in the client module's appropriate package:

{service}/{service}-client/src/main/java/com/salesforce/multicloudj/{service}/driver/

Follow existing patterns:

• Request objects use Builder pattern with Builder inner class
• Response objects are immutable with getters
• Use @Getter from Lombok where the module already uses it

3b. Abstract Driver Method

Add the abstract do* method to the abstract class (e.g., AbstractBlobStore):

protected abstract {ResponseType} do{OperationName}({RequestType} request);

3c. Public Client Method

Add the public method to the client class (e.g., BucketClient):

public {ResponseType} {operationName}({RequestType} request) {
    return multiCloudJLogger.traceOperation(
        "{operationName}",
        () -> blobStore.{operationName}(request)
    );
}

3d. Validation

Add validation rules to the validator class (e.g., BlobStoreValidator) if inputs need validation.

Present the API design to the user for approval before implementing providers.

Phase 4: Provider Implementations

For each provider (aws, gcp, ali):

4a. Implement the do* Method

In the provider's main class (e.g., AwsBlobStore):

@Override
protected {ResponseType} do{OperationName}({RequestType} request) {
    // 1. Transform multicloudj request → provider SDK request
    // 2. Call provider SDK
    // 3. Transform provider SDK response → multicloudj response
    // 4. Handle provider-specific exceptions via getException()
}

4b. Add Transformer Methods

In the provider's transformer class (e.g., AwsTransformer):

• to{ProviderRequest}({MulticloudRequest}) - converts outgoing request
• to{MulticloudResponse}({ProviderResponse}) - converts incoming response

4c. Exception Mapping

Update ErrorCodeMapping (or equivalent) if the new operation introduces new error conditions.

4d. Build and Verify

mvn clean install -DskipTests
mvn test -pl {service}/{service}-{provider}

Phase 5: Unit Tests

Write unit tests that verify real behavior, not fake coverage.

What Makes a Real Unit Test

• Tests behavior, not implementation - verify the output given an input
• Mocks the cloud SDK client (S3Client, Storage, etc.), NOT internal classes
• Covers error paths - exception mapping, validation failures, edge cases
• Tests transformer logic - correct field mapping between types

What Is Fake Coverage (DO NOT WRITE)

• Tests that only verify a method was called without checking correctness
• Tests that pass null/empty and assert no exception (unless that's the contract)
• Tests that duplicate the implementation logic in assertions
• Tests with no meaningful assertions

Run Unit Tests

mvn clean install -DskipTests
mvn test -pl {service}/{service}-aws -Dtest={TestClassName}
mvn test -pl {service}/{service}-gcp -Dtest={TestClassName}

Phase 6: Conformance Tests

Conformance tests ensure all providers behave identically for the same operation.

6a. Abstract Conformance Test

Add test methods to the existing abstract IT class in the client module (e.g., AbstractBlobStoreIT):

{service}/{service}-client/src/test/java/.../Abstract{Service}StoreIT.java

Every conformance test must include both positive and negative scenarios:

• Positive scenario: Happy path — the operation succeeds with valid inputs and the result is correct
• Negative scenario: Error path — the operation fails gracefully with invalid inputs (e.g., nonexistent key, invalid parameters) and throws the expected exception

This ensures all providers handle both success and failure consistently.

6b. Run in Record Mode (Local with Credentials)

Record mode captures real HTTP interactions with the cloud provider.

Ask the user for credentials:

• AWS: AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_SESSION_TOKEN
• GCP: GOOGLE_APPLICATION_CREDENTIALS (path to service account JSON)
• Alibaba: Handled separately on special machines - skip record mode for Ali and override in the Ali{Service}IT and add @Disabled so they are not executed.

Only run the specific tests impacted by your change - never record the entire IT suite. This keeps the PR minimal and avoids regenerating unrelated mapping files.

# AWS record - only the new/changed test method
export AWS_ACCESS_KEY_ID="{from_user}"
export AWS_SECRET_ACCESS_KEY="{from_user}"
export AWS_SESSION_TOKEN="{from_user}"
mvn test -pl {service}/{service}-aws -Dtest="{Provider}{Service}StoreIT#{testMethodName}" -Drecord

# GCP record - only the new/changed test method
export GOOGLE_APPLICATION_CREDENTIALS="{path_from_user}"
mvn test -pl {service}/{service}-gcp -Dtest="{Provider}{Service}StoreIT#{testMethodName}" -Drecord

After recording:

• Verify new mapping files appear in src/test/resources/mappings/
• Verify no sensitive credentials leaked into mapping files
• NEVER manually edit or create mapping files - all mappings must be generated by WireMock recording. If stale files exist from a failed run, delete them and re-record.
• Commit the mapping files (they enable CI replay mode)

6c. Run in Replay Mode (CI - Default)

# No credentials needed - uses recorded WireMock stubs
mvn test -pl {service}/{service}-aws -Dtest="{Provider}{Service}StoreIT#{testMethodName}"
mvn test -pl {service}/{service}-gcp -Dtest="{Provider}{Service}StoreIT#{testMethodName}"

6d. Alibaba Conformance Tests

Write the abstract conformance test (it runs for all providers). The Ali-specific IT class extends it just like AWS/GCP, but recording happens on dedicated machines. The IT harness class already exists (e.g., AliBlobStoreIT) — do not attempt to record locally for Ali.

Phase 7: Create Pull Request

After implementation and testing are complete, create a PR following MultiCloudJ conventions.

7a. Commit Changes

# Stage all changes
git add -A

# Create commit with descriptive message
git commit -m "{service}: {brief description of feature}

{Detailed description of what was changed and why}

- Added X to Y
- Updated Z for compatibility
- Tests: {description of tests added}
"

7b. Push Branch

# Push to origin
git push origin {branch-name}

7c. Create PR with Proper Title

PR Title Format: {service}: {concise description}

Examples:

• sts: onboard proxy configurations for sts interface
• blob: add object versioning support
• docstore: implement batch delete with filters
• pubsub: add dead-letter queue configuration

PR Description Template:

## Summary
{1-2 sentence overview of the feature}

## Changes
- Added `{MethodName}` to `{ClassName}`
- Implemented {feature} for AWS, GCP, and Alibaba Cloud
- Added unit tests for {components}
- {If applicable} Added conformance tests with WireMock recording

## API Example
```java
// Show how users will call the new API
{ServiceClient} client = {ServiceClient}.builder("aws")
    .withRegion("us-west-2")
    .{newMethod}({parameters})
    .build();

Testing

• Unit tests: mvn test -pl {service}/{service}-{provider}
• {If applicable} Conformance tests: recorded for AWS/GCP, replay mode verified

Cross-Cloud Compatibility

• AWS: {describe implementation or limitations}
• GCP: {describe implementation or limitations}
• Alibaba: {describe implementation or limitations}

### 7d. Verify PR Checklist

Before submitting, verify:
- [ ] PR title follows `{service}: {description}` format
- [ ] All unit tests pass locally
- [ ] Conformance tests pass (if applicable)
- [ ] Checkstyle passes (Google Java Style)
- [ ] No provider-specific code in `-client` modules
- [ ] Documentation/examples added (if needed)
- [ ] WireMock mappings committed (if recorded)
- [ ] No credentials leaked in mapping files

## File Checklist

For a feature added to an existing service (e.g., blob), you will typically touch:

| Layer | File | Change |
|-------|------|--------|
| Request/Response | `{service}-client/.../driver/{Request}.java` | New file |
| Request/Response | `{service}-client/.../driver/{Response}.java` | New file |
| Abstract driver | `{service}-client/.../driver/Abstract{Store}.java` | Add `do*` method |
| Public API | `{service}-client/.../client/{Client}.java` | Add public method |
| Validator | `{service}-client/.../driver/{Validator}.java` | Add validation (if needed) |
| AWS impl | `{service}-aws/.../Aws{Store}.java` | Implement `do*` |
| AWS transformer | `{service}-aws/.../AwsTransformer.java` | Add conversion methods |
| GCP impl | `{service}-gcp/.../Gcp{Store}.java` | Implement `do*` |
| GCP transformer | `{service}-gcp/.../GcpTransformer.java` | Add conversion methods |
| Ali impl | `{service}-ali/.../Ali{Store}.java` | Implement `do*` |
| Ali transformer | `{service}-ali/.../AliTransformer.java` | Add conversion methods |
| AWS unit test | `{service}-aws/.../{FeatureOrComponent}Test.java` | New or updated |
| GCP unit test | `{service}-gcp/.../{FeatureOrComponent}Test.java` | New or updated |
| Ali unit test | `{service}-ali/.../{FeatureOrComponent}Test.java` | New or updated |
| Conformance test | `{service}-client/.../Abstract{Service}StoreIT.java` | Add test methods |
| AWS IT harness | `{service}-aws/.../Aws{Service}StoreIT.java` | Already exists |
| GCP IT harness | `{service}-gcp/.../Gcp{Service}StoreIT.java` | Already exists |
| Ali IT harness | `{service}-ali/.../Ali{Service}StoreIT.java` | Already exists |
| AWS mappings | `{service}-aws/src/test/resources/mappings/` | Recorded stubs |
| GCP mappings | `{service}-gcp/src/test/resources/mappings/` | Recorded stubs |

## Common Mistakes

- **Adding provider-specific types in client module** - the client module must never import AWS/GCP/Ali SDKs
- **Skipping the user interview** - assumptions about semantics lead to rework
- **Implementing before researching all three clouds** - discovering a provider can't support the feature after implementation is costly
- **Writing conformance tests that are provider-specific** - the abstract IT must work for ALL providers
- **Leaking credentials in WireMock recordings** - always verify mapping files before committing
- **Writing unit tests that only test happy path** - error mapping and edge cases are where bugs hide
- **Not running `mvn clean install -DskipTests` before running individual module tests** - inter-module dependencies need to be built first
- **Manually editing or creating WireMock mapping files** - all mappings must be produced by WireMock recording only. If a recording is stale or corrupt, delete it and re-record.
- **Recording the entire IT suite instead of just the new test** - only run the specific impacted test method in record mode to keep the PR diff minimal

## Red Flags - STOP and Reconsider

- You're implementing a provider without having researched how the other providers handle it
- You're adding a method to the abstract class without a corresponding public client method
- You're writing a conformance test that uses provider-specific setup
- You're importing `software.amazon.awssdk` or `com.google.cloud` in a `-client` module
- You're writing a unit test with no assertions about returned values or thrown exceptions
- You're running record mode without asking the user for credentials first
- You're about to manually edit a WireMock mapping JSON file - STOP: delete and re-record instead
- You're running the full IT class in record mode when only one test method is new/changed - use `#testMethodName` filter
- **You're writing a comment in a provider that references another provider** (e.g., "Unlike GCP", "Similar to AWS") - STOP: providers must be isolated
- **You're importing classes from another provider module** (e.g., `blob-aws` importing from `blob-gcp`) - STOP: providers cannot depend on each other